bandrami

Copyright infringement is in most cases tortious rather than criminal.

Gorilla also makes one though even they seemed surprised when I ordered some a few months ago

Poe

Why not just post the prompt you used? I have access to LLMs too.

> you are doing that for LLM code right?

Existentially weary infosec guy laughter

We had to move my non-ambulatory[1] father from Mississippi to Virginia a couple of months ago. The vehicle probably didn't have all the staff and equipment of an ambulance but it did have an RN, a crash cart, and some other expensive crap that goes "ping". The cost was $3K for an 800 mile trip (slightly more because I included the "snacks and a DVD player" package).

[1] Oh, holy crap, I just got why ambulances are called that

This is a pendulum I've now seen fully swing twice since I enlisted 30 years ago.

"We need more integrated logistics because the teeth can't fight without the tail!"

Some years pass

"Why do we have all these non-combat roles in the military? Shrink everything down and focus on warfighting!"

More years pass

"Why can't we do any support internally? We need stronger and more integrated logistics!"

Lather, rinse, repeat.

I was very technically impressed with xz, FWIW.

Note that the troublesome vector isn't code execution by the LLM, it's instructions to produce vulnerable code

Did you ever read the paper on how Ken Thompson backdoored the original C compiler?

My view is that the same people who pulled off xz are putting God knows what in the weights

> it’s Chinese and backed by god knows who, and no one cares

Existentially weary infosec guy voice I promise you some of us care, and the rest of you will realize why in a year or so

What if it's a domain I just don't want anybody to use?

> HTML at the time was intended as an application of SGML

Worse, it was an extended superset (ha!) of SGML. At least 20 years ago, SGML::Parser would reject some valid HTML documents.

That said, it was really easy to type correctly in a text editor (especially compared to actual SGML), particularly one that indented and matched tags for you.

What were these "troublesome parts"? The whole point of HTML's design is that it's incredibly easy for a human to write correctly.

Inference providers have been getting a firehose of investor cash to keep the chips running (and are looking around very nervously as that firehose starts to sputter).

Paging Mr. Son. Mr. Son, please pick up line 3.

All "density" means is that it's a quotient; I use spectral densities daily and that's "count per Hertz" which weirdly enough works out in normal units to be seconds.

I lost count of the number of times I've seen a junior dev call qsort on an array that is guaranteed to only have a few hundred members at most.

> Note that whoever you buy from will be able to read all your tokens, so don’t use it for anything confidential/financial.

Note that without an enterprise agreement, both OpenAI and Anthropic make no commitments as to what they will and won't do with your data either.

> Based on what?

Based on the money they're spending on compute being less than the money they're getting from customers.

No I mean Anthropic has only claimed a profitable quarter based on xAI giving them two months of free compute, and both Anthropic and OpenAI are counting discounted revenue as actual revenue. They haven't found a way to sell inference for less than it costs them yet, and when they tried earlier this quarter their customers bailed.

If they had high margins they wouldn't be issuing senior debt with a 18.5% coupon payment (and failing to fully subscribe it), nor would they need Elon to give them two months of free compute in order to appear profitable for a single quarter.

Literally nobody is making money on inference

Sure, and Anthropic is allowed to cancel the licenses of people doing that, as they do

The relief available to a licensor for violating a license use restriction is cancellation of the license. And they're free to do that, just like Alibaba is free to pay somebody in Hyderabad $20 to make another one.

DMCA can't apply in this case because (this is the "C" in its initialism) it is based on copyright protections, which the output of Claude is not eligible for.

My long-term prediction for the sector is that frontier models will be so expensive that they will only be available for grant-funded projects at research institutions, like supercomputer clusters were 25 years ago.

No, it's part of the capability theft. They resell Claude tokens cheaply and then simultaneously log everything for distillation. Even if they take a small loss on the token sales it's much cheaper than the equivalent compute.

The current case law in the US is that the raw output of an LLM cannot be copyrighted without further meaningful arrangement or alteration by a human author.

Free over-the-air network TV is (generally) copyrighted.

The output of LLMs cannot be copyrighted. This isn't a semantic game; it's literally the case that Anthropic cannot seek relief for people duplicating the output of an LLM.

If it's streaming an uncopyrightable product, absolutely. This isn't even a gray area.

I'm curious why you think you cannot re-stream a public domain stream.